Models

BlockRun provides access to models from multiple providers through a unified API.

List Models

GET https://blockrun.ai/api/v1/models

Returns a list of available models with pricing information. The response now includes extended metadata for each model.

Response Fields

Each model object in the response includes:

Only models that are currently available are returned — there is no available flag to check.

Example Response

{
  "object": "list",
  "data": [
    {
      "id": "openai/gpt-5.5",
      "object": "model",
      "created": 1700000000,
      "owned_by": "openai",
      "name": "GPT-5.5",
      "description": "OpenAI's flagship — first fully retrained base since GPT-4.5; 1M context, 128K output, native agent + computer use",
      "context_window": 1050000,
      "max_output": 128000,
      "categories": ["chat", "coding", "vision"],
      "billing_mode": "paid",
      "pricing": { "input": 5.00, "output": 30.00 }
    }
  ]
}

Available Models (chat / image / video / music / speech / sound effects)

All prices shown are provider rates — and, for per-token chat, also the billed rates: BlockRun adds no platform margin on chat tokens, only the flat $0.001/request transaction fee. Media and Live Search still carry a 5% platform fee.

OpenAI GPT-6 Family

Astra is the flagship; Sol (released 2026-09-22 with Luna) is the cost-efficient tier below it, and GPT-6.1 Sol (released 2026-09-29) is its refresh at the same price with a cheaper cached rate ($0.10/M against Sol's $0.20/M). All four bill a long-context tier — 2x input and cached rates, 1.5x output for the whole request once the prompt exceeds 272K tokens (mirrors OpenAI's official pricing: gpt-6-sol and gpt-6.1-sol are $4.00/M in · $15.00/M out above the threshold). Astra, Sol and Luna serve Flex at half the standard rate; GPT-6.1 Sol does not offer Flex yet.

GPT-6 models accept reasoning_effort low through max and reject temperature other than 1, top_p, penalties, logprobs and stop. On plain chat the gateway lowers max to xhigh. Tool calls, which it serves through OpenAI's Responses API, run at the effort you ask for, max included, and at medium when none is sent. Higher effort spends more reasoning, which bills as output, so the per-call price scales with it: the output share of the quote is 2.5x at high, 3x at xhigh and 4.5x at max (never past max_tokens); no effort, low and medium are priced as before. none and minimal are raised to low on Astra and GPT-6.1 Sol; Sol and Luna accept none, and raise minimal to low.

OpenAI GPT-5.6 Family

Released 2026-07-09 — three fixed tiers (Sol / Terra / Luna) replacing the single-model-plus-effort-knob line. Note: on /v1/chat/completions, GPT-5.6 accepts function tools only with reasoning_effort: "none" — the BlockRun gateway sets this automatically when tools are present, so no client change is needed.

OpenAI GPT-5.5 Family

Released 2026-04-23 — first fully retrained base since GPT-4.5.

OpenAI GPT-5.4 Family

OpenAI GPT-5 Family

OpenAI O-Series (Reasoning)

OpenAI GPT-4.1 / GPT-4o

Anthropic Claude

Google Gemini

Gemini Pro models (gemini-2.5-pro, gemini-3.1-pro) bill a long-context tier — 2x input, 1.5x output above 200K prompt tokens (mirrors Google's official pricing: gemini-2.5-pro is $2.50/M in · $15.00/M out, gemini-3.1-pro is $4.00/M in · $18.00/M out above the threshold). Flash / Flash-Lite are flat-priced.

xAI Grok

Grok bills a long-context tier at 2x the rates above once a request's prompt reaches 200K tokens (mirrors xAI's official pricing — e.g. Grok 4.7 is $4.00/M in · $12.00/M out above the threshold). Live Search adds one flat surcharge per call: $0.30 on Grok 4.5 / 4.6 / 4.7 / 4.20, $0.12 on Grok 4.3 and below. Grok Imagine image/video SKUs are listed under Image / Video Generation below.

DeepSeek

Z.AI

Moonshot

K3 is the current flagship — a 2.8-trillion-parameter open MoE with a 1M-token context window, image + text input, returning reasoning_content on completions. K2.7 (256K, adds video input) remains routable but superseded; K2.6 and K2.5 are earlier multi-modal generations, still routable.

MiniMax

Qwen (Alibaba)

Tencent

Poolside

Mistral

Xiaomi

Flat-priced (open-weight)

Open-weight models billed at a flat per-request price instead of per token — the request is the unit, whatever its length.

Free Tier (open-weight)

Open-weight models served free of charge (no x402 payment), subject to a small per-IP rate limit. The free tier auto-routes around any temporarily unavailable model, so the live set is best read from GET /api/v1/models (filter on billing_mode: "free"). There are 8 free models listed on mainnet.

Image Generation

Media prices below include the 5% media margin; the flat $0.001 transaction fee is added per call.

Video Generation

Seedance defaults to 720p with synced audio for text-to-video; pass resolution / generate_audio to override. See Video Generation API.

For character consistency across multiple Seedance videos, enroll a Virtual Portrait ($0.011 one-time, no government ID) for AI characters, or a RealFace ($0.011 one-time, no government ID, requires brief on-phone liveness check) for real people. Pass the returned ta_xxx as real_face_asset_id.

Model Categories

Each model includes a categories array in the API response. Categories indicate model capabilities:

• chat - General conversation
• reasoning - Complex problem-solving
• coding - Code generation and analysis
• vision - Image understanding

Filter models by category:

models = client.list_models()
reasoning_models = [m for m in models if "reasoning" in m.get("categories", [])]

Pricing

Prices are per 1 million tokens. Your actual cost depends on:

1. Input tokens - Length of your prompt and context
2. Output tokens - Length of the model's response
3. Transaction fee - a flat $0.001 per request; chat tokens carry no platform margin (media and Live Search carry 5%)

The SDK calculates the exact price before each request.

Want to save 84% automatically? ClawRouter routes each request to the cheapest model that can handle it.

Example

What's next?